package executor import (	"bufio"	"bytes"	"compress/flate"	"compress/gzip"	"context"	"fmt"	"io"	"net/http"	"strings"	"time" 	"github.com/andybalholm/brotli"	"github.com/klauspost/compress/zstd"	claudeauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"	log "github.com/sirupsen/logrus"	"github.com/tidwall/gjson"	"github.com/tidwall/sjson" 	"github.com/gin-gonic/gin") // ClaudeExecutor is a stateless executor for Anthropic Claude over the messages API.// If api_key is unavailable on auth, it falls back to legacy via ClientAdapter.type ClaudeExecutor struct {	cfg *config.Config} func NewClaudeExecutor(cfg *config.Config) *ClaudeExecutor { return &ClaudeExecutor{cfg: cfg} } func (e *ClaudeExecutor) Identifier() string { return "claude" } func (e *ClaudeExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil } func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {	apiKey, baseURL := claudeCreds(auth) 	if baseURL == "" {		baseURL = "https://api.anthropic.com"	}	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)	defer reporter.trackFailure(ctx, &err)	from := opts.SourceFormat	to := sdktranslator.FromString("claude")	// Use streaming translation to preserve function calling, except for claude.	stream := from != to	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)	if upstreamModel == "" {		upstreamModel = req.Model	}	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {		upstreamModel = modelOverride	} else if !strings.EqualFold(upstreamModel, req.Model) {		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {			upstreamModel = modelOverride		}	}	body, _ = sjson.SetBytes(body, "model", upstreamModel)	// Inject thinking config based on model metadata for thinking variants	body = e.injectThinkingConfig(req.Model, req.Metadata, body) 	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {		body = checkSystemInstructions(body)	}	body = applyPayloadConfig(e.cfg, req.Model, body) 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)	body = disableThinkingIfToolChoiceForced(body) 	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled	body = ensureMaxTokensForThinking(req.Model, body) 	// Extract betas from body and convert to header	var extraBetas []string	extraBetas, body = extractAndRemoveBetas(body) 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))	if err != nil {		return resp, err	}	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)	var authID, authLabel, authType, authValue string	if auth != nil {		authID = auth.ID		authLabel = auth.Label		authType, authValue = auth.AccountInfo()	}	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{		URL:       url,		Method:    http.MethodPost,		Headers:   httpReq.Header.Clone(),		Body:      body,		Provider:  e.Identifier(),		AuthID:    authID,		AuthLabel: authLabel,		AuthType:  authType,		AuthValue: authValue,	}) 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)	httpResp, err := httpClient.Do(httpReq)	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		return resp, err	}	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {		b, _ := io.ReadAll(httpResp.Body)		appendAPIResponseChunk(ctx, e.cfg, b)		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))		err = statusErr{code: httpResp.StatusCode, msg: string(b)}		if errClose := httpResp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		return resp, err	}	decodedBody, err := decodeResponseBody(httpResp.Body, httpResp.Header.Get("Content-Encoding"))	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		if errClose := httpResp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		return resp, err	}	defer func() {		if errClose := decodedBody.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}	}()	data, err := io.ReadAll(decodedBody)	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		return resp, err	}	appendAPIResponseChunk(ctx, e.cfg, data)	if stream {		lines := bytes.Split(data, []byte("\n"))		for _, line := range lines {			if detail, ok := parseClaudeStreamUsage(line); ok {				reporter.publish(ctx, detail)			}		}	} else {		reporter.publish(ctx, parseClaudeUsage(data))	}	var param any	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, data, &param)	resp = cliproxyexecutor.Response{Payload: []byte(out)}	return resp, nil} func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {	apiKey, baseURL := claudeCreds(auth) 	if baseURL == "" {		baseURL = "https://api.anthropic.com"	}	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)	defer reporter.trackFailure(ctx, &err)	from := opts.SourceFormat	to := sdktranslator.FromString("claude")	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)	if upstreamModel == "" {		upstreamModel = req.Model	}	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {		upstreamModel = modelOverride	} else if !strings.EqualFold(upstreamModel, req.Model) {		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {			upstreamModel = modelOverride		}	}	body, _ = sjson.SetBytes(body, "model", upstreamModel)	// Inject thinking config based on model metadata for thinking variants	body = e.injectThinkingConfig(req.Model, req.Metadata, body)	body = checkSystemInstructions(body)	body = applyPayloadConfig(e.cfg, req.Model, body) 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)	body = disableThinkingIfToolChoiceForced(body) 	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled	body = ensureMaxTokensForThinking(req.Model, body) 	// Extract betas from body and convert to header	var extraBetas []string	extraBetas, body = extractAndRemoveBetas(body) 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))	if err != nil {		return nil, err	}	applyClaudeHeaders(httpReq, auth, apiKey, true, extraBetas)	var authID, authLabel, authType, authValue string	if auth != nil {		authID = auth.ID		authLabel = auth.Label		authType, authValue = auth.AccountInfo()	}	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{		URL:       url,		Method:    http.MethodPost,		Headers:   httpReq.Header.Clone(),		Body:      body,		Provider:  e.Identifier(),		AuthID:    authID,		AuthLabel: authLabel,		AuthType:  authType,		AuthValue: authValue,	}) 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)	httpResp, err := httpClient.Do(httpReq)	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		return nil, err	}	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {		b, _ := io.ReadAll(httpResp.Body)		appendAPIResponseChunk(ctx, e.cfg, b)		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))		if errClose := httpResp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		err = statusErr{code: httpResp.StatusCode, msg: string(b)}		return nil, err	}	decodedBody, err := decodeResponseBody(httpResp.Body, httpResp.Header.Get("Content-Encoding"))	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		if errClose := httpResp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		return nil, err	}	out := make(chan cliproxyexecutor.StreamChunk)	stream = out	go func() {		defer close(out)		defer func() {			if errClose := decodedBody.Close(); errClose != nil {				log.Errorf("response body close error: %v", errClose)			}		}() 		// If from == to (Claude → Claude), directly forward the SSE stream without translation		if from == to {			scanner := bufio.NewScanner(decodedBody)			scanner.Buffer(nil, 52_428_800) // 50MB			for scanner.Scan() {				line := scanner.Bytes()				appendAPIResponseChunk(ctx, e.cfg, line)				if detail, ok := parseClaudeStreamUsage(line); ok {					reporter.publish(ctx, detail)				}				// Forward the line as-is to preserve SSE format				cloned := make([]byte, len(line)+1)				copy(cloned, line)				cloned[len(line)] = '\n'				out <- cliproxyexecutor.StreamChunk{Payload: cloned}			}			if errScan := scanner.Err(); errScan != nil {				recordAPIResponseError(ctx, e.cfg, errScan)				reporter.publishFailure(ctx)				out <- cliproxyexecutor.StreamChunk{Err: errScan}			}			return		} 		// For other formats, use translation		scanner := bufio.NewScanner(decodedBody)		scanner.Buffer(nil, 52_428_800) // 50MB		var param any		for scanner.Scan() {			line := scanner.Bytes()			appendAPIResponseChunk(ctx, e.cfg, line)			if detail, ok := parseClaudeStreamUsage(line); ok {				reporter.publish(ctx, detail)			}			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(line), &param)			for i := range chunks {				out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}			}		}		if errScan := scanner.Err(); errScan != nil {			recordAPIResponseError(ctx, e.cfg, errScan)			reporter.publishFailure(ctx)			out <- cliproxyexecutor.StreamChunk{Err: errScan}		}	}()	return stream, nil} func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {	apiKey, baseURL := claudeCreds(auth) 	if baseURL == "" {		baseURL = "https://api.anthropic.com"	} 	from := opts.SourceFormat	to := sdktranslator.FromString("claude")	// Use streaming translation to preserve function calling, except for claude.	stream := from != to	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)	if upstreamModel == "" {		upstreamModel = req.Model	}	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {		upstreamModel = modelOverride	} else if !strings.EqualFold(upstreamModel, req.Model) {		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {			upstreamModel = modelOverride		}	}	body, _ = sjson.SetBytes(body, "model", upstreamModel) 	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {		body = checkSystemInstructions(body)	} 	// Extract betas from body and convert to header (for count_tokens too)	var extraBetas []string	extraBetas, body = extractAndRemoveBetas(body) 	url := fmt.Sprintf("%s/v1/messages/count_tokens?beta=true", baseURL)	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))	if err != nil {		return cliproxyexecutor.Response{}, err	}	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)	var authID, authLabel, authType, authValue string	if auth != nil {		authID = auth.ID		authLabel = auth.Label		authType, authValue = auth.AccountInfo()	}	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{		URL:       url,		Method:    http.MethodPost,		Headers:   httpReq.Header.Clone(),		Body:      body,		Provider:  e.Identifier(),		AuthID:    authID,		AuthLabel: authLabel,		AuthType:  authType,		AuthValue: authValue,	}) 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)	resp, err := httpClient.Do(httpReq)	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		return cliproxyexecutor.Response{}, err	}	recordAPIResponseMetadata(ctx, e.cfg, resp.StatusCode, resp.Header.Clone())	if resp.StatusCode < 200 || resp.StatusCode >= 300 {		b, _ := io.ReadAll(resp.Body)		appendAPIResponseChunk(ctx, e.cfg, b)		if errClose := resp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		return cliproxyexecutor.Response{}, statusErr{code: resp.StatusCode, msg: string(b)}	}	decodedBody, err := decodeResponseBody(resp.Body, resp.Header.Get("Content-Encoding"))	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		if errClose := resp.Body.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}		return cliproxyexecutor.Response{}, err	}	defer func() {		if errClose := decodedBody.Close(); errClose != nil {			log.Errorf("response body close error: %v", errClose)		}	}()	data, err := io.ReadAll(decodedBody)	if err != nil {		recordAPIResponseError(ctx, e.cfg, err)		return cliproxyexecutor.Response{}, err	}	appendAPIResponseChunk(ctx, e.cfg, data)	count := gjson.GetBytes(data, "input_tokens").Int()	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)	return cliproxyexecutor.Response{Payload: []byte(out)}, nil} func (e *ClaudeExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {	log.Debugf("claude executor: refresh called")	if auth == nil {		return nil, fmt.Errorf("claude executor: auth is nil")	}	var refreshToken string	if auth.Metadata != nil {		if v, ok := auth.Metadata["refresh_token"].(string); ok && v != "" {			refreshToken = v		}	}	if refreshToken == "" {		return auth, nil	}	svc := claudeauth.NewClaudeAuth(e.cfg)	td, err := svc.RefreshTokens(ctx, refreshToken)	if err != nil {		return nil, err	}	if auth.Metadata == nil {		auth.Metadata = make(map[string]any)	}	auth.Metadata["access_token"] = td.AccessToken	if td.RefreshToken != "" {		auth.Metadata["refresh_token"] = td.RefreshToken	}	auth.Metadata["email"] = td.Email	auth.Metadata["expired"] = td.Expire	auth.Metadata["type"] = "claude"	now := time.Now().Format(time.RFC3339)	auth.Metadata["last_refresh"] = now	return auth, nil} // extractAndRemoveBetas extracts the "betas" array from the body and removes it.// Returns the extracted betas as a string slice and the modified body.func extractAndRemoveBetas(body []byte) ([]string, []byte) {	betasResult := gjson.GetBytes(body, "betas")	if !betasResult.Exists() {		return nil, body	}	var betas []string	if betasResult.IsArray() {		for _, item := range betasResult.Array() {			if s := strings.TrimSpace(item.String()); s != "" {				betas = append(betas, s)			}		}	} else if s := strings.TrimSpace(betasResult.String()); s != "" {		betas = append(betas, s)	}	body, _ = sjson.DeleteBytes(body, "betas")	return betas, body} // injectThinkingConfig adds thinking configuration based on metadata using the unified flow.// It uses util.ResolveClaudeThinkingConfig which internally calls ResolveThinkingConfigFromMetadata// and NormalizeThinkingBudget, ensuring consistency with other executors like Gemini.func (e *ClaudeExecutor) injectThinkingConfig(modelName string, metadata map[string]any, body []byte) []byte {	budget, ok := util.ResolveClaudeThinkingConfig(modelName, metadata)	if !ok {		return body	}	return util.ApplyClaudeThinkingConfig(body, budget)} // disableThinkingIfToolChoiceForced checks if tool_choice forces tool use and disables thinking.// Anthropic API does not allow thinking when tool_choice is set to "any" or a specific tool.// See: https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#important-considerationsfunc disableThinkingIfToolChoiceForced(body []byte) []byte {	toolChoiceType := gjson.GetBytes(body, "tool_choice.type").String()	// "auto" is allowed with thinking, but "any" or "tool" (specific tool) are not	if toolChoiceType == "any" || toolChoiceType == "tool" {		// Remove thinking configuration entirely to avoid API error		body, _ = sjson.DeleteBytes(body, "thinking")	}	return body} // ensureMaxTokensForThinking ensures max_tokens > thinking.budget_tokens when thinking is enabled.// Anthropic API requires this constraint; violating it returns a 400 error.// This function should be called after all thinking configuration is finalized.// It looks up the model's MaxCompletionTokens from the registry to use as the cap.func ensureMaxTokensForThinking(modelName string, body []byte) []byte {	thinkingType := gjson.GetBytes(body, "thinking.type").String()	if thinkingType != "enabled" {		return body	} 	budgetTokens := gjson.GetBytes(body, "thinking.budget_tokens").Int()	if budgetTokens <= 0 {		return body	} 	maxTokens := gjson.GetBytes(body, "max_tokens").Int() 	// Look up the model's max completion tokens from the registry

Lines 1–500 of 783. Load more